Coincidentally enough, we just got through writing a nest of regexps to retrospectively validate about 2K user entered addresses, and they turned out to be surp

Coincidentally enough, we just got through writing a nest of regexps to retrospectively validate about 2K user entered addresses, and they turned out to be surprisingly well entered, probably about 98% parsable. In this case the user was entering a field called "street" and we were validating down to a number followed by a street type (e.g ST) and a few local wrinkles such as additions like "RD1" for rural addresses. The most common error, which we brute force corrected, were various misspellings of "crescent". We did not however attempt to unpack the street number, which fortunately was not required.